Skip to content

faith-ogun/CareSignal

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CareSignal

The ambient scribe for domiciliary care. We listen so carers don't have to type.

CareSignal is an ambient AI care scribe for UK domiciliary (home) care. A carer activates the app at the start of a home visit, and CareSignal listens, transcribes, and structures the conversation into a complete care note with 10 structured fields mapped to CQC's five inspection domains, ready for review and submission in under 60 seconds.

Built for HSIL Harvard Health Systems Hackathon 2026 (Challenge #6: Shortage of Healthcare Workers) at UCL Hub, London.


Six Pillars

Pillar Description Status
Ambient AI Scribe On-device audio capture + Gemini 2.5 Flash extracts 10 structured care fields from natural conversation in 9 languages Built
Pre-Visit AI Brief Auto-generated brief from visit history with flags, care plan changes, patient preferences, and ElevenLabs voice narration Built
Lone Worker Safety One-tap emergency: GPS + Google Maps location link to agency contact with Twilio SMS backup Built
Manager Dashboard Visit feed, calendar view, analytics, anomaly flags, CQC-ready PDF report export, quick messaging Built
Notifications & Comms Pre-visit reminder emails (Resend), district nurse insights summaries, in-app chat persisted to Firestore Built
Care System Integrations Export to Birdie, PASS (everyLIFE), generic NHS Digital format, and FHIR R4 Bundle with SNOMED CT coding Built

10 Structured Care Fields

Each field maps to CQC's five inspection domains (Safe, Effective, Caring, Responsive, Well-Led):

Field CQC Domain Example
Fluid Intake Safe 450ml
Food Intake Effective Half portion of soup
Mood Caring Cheerful, engaged
Mobility Safe Walked unaided
Personal Care Effective Wash completed
Medication Safe Morning tablets given
Pain Level Caring Mild, 3/10
Wellbeing Caring Seemed lonely
Carer Concerns Responsive Bruising on arm
Visit Duration Well-Led 28 min

Tech Stack

Frontend

  • React 18 + TypeScript + Vite
  • Tailwind CSS v4 with custom design tokens
  • Framer Motion for animations
  • Recharts for analytics charts
  • jsPDF for CQC report export
  • React Router v6

Backend

  • Node.js + Express
  • Google Cloud Speech-to-Text (9 languages, enhanced model)
  • Google Gemini 2.5 Flash (structured field extraction, pre-visit briefs)
  • ElevenLabs (text-to-speech, Sarah voice, turbo v2.5 model)
  • Twilio (emergency SMS + pre-visit SMS reminders)
  • Resend (transactional email with styled HTML templates)
  • Firebase Admin SDK + Firestore

Infrastructure

  • GCP project: caresignal-hsil (europe-west2 / London)
  • Firestore: native mode, europe-west2
  • Privacy-by-design: audio deleted post-transcription, transcripts are transient, only structured fields persist
  • TLS 1.3 in transit, AES-256 at rest

Getting Started

Prerequisites

  • Node.js 18+
  • Google Cloud CLI (gcloud) authenticated
  • GCP project with Speech-to-Text, Firestore, and Generative Language APIs enabled

Setup

# Clone
git clone https://github.com/faith-ogun/CareSignal.git
cd CareSignal

# Backend
cd backend
npm install
cp .env.example .env  # Edit with your API keys
npm start             # Runs on http://localhost:3001

# Frontend (new terminal)
cd frontend
npm install
npm run dev           # Runs on http://localhost:5173

Environment Variables (backend/.env)

# Google Cloud
GEMINI_API_KEY=your_gemini_api_key
GEMINI_MODEL=gemini-2.5-flash
GCP_PROJECT_ID=caresignal-hsil

# ElevenLabs (TTS)
ELEVENLABS_API_KEY=your_elevenlabs_key

# Twilio (SMS)
TWILIO_ACCOUNT_SID=your_twilio_sid
TWILIO_AUTH_TOKEN=your_twilio_token
TWILIO_PHONE_NUMBER=+1...
EMERGENCY_CONTACT=+44...

# Resend (Email)
RESEND_API_KEY=your_resend_key
CARER_EMAIL=carer@example.com

PORT=3001

API Endpoints

Method Endpoint Description
GET /api/health Service status check
GET /api/languages List supported transcription languages
GET /api/visits Get completed visits (Firestore)
POST /api/visits Save completed visit (Firestore)
GET /api/alerts Get emergency alerts (Firestore)
POST /api/alerts Save emergency alert (Firestore)
POST /api/transcribe Audio file upload to transcript (multilingual)
POST /api/transcribe-base64 Base64 audio to transcript (multilingual)
POST /api/extract Transcript to 10 structured care fields
POST /api/emergency Lone worker emergency SMS alert
POST /api/pre-visit-brief Generate pre-visit brief from history
POST /api/tts Text to speech via ElevenLabs
POST /api/send-reminder Pre-visit reminder (email + SMS)
POST /api/send-summary District nurse insights summary email
GET /api/messages Get chat messages (Firestore)
POST /api/messages Save chat message (Firestore)

Multilingual Support

CareSignal supports 9 languages for speech-to-text transcription, covering the most spoken languages in the UK care sector:

Language Code Use Case
English (UK) en-GB Default
Polish pl-PL Largest non-English group in UK
Punjabi pa-IN Second most spoken South Asian language
Urdu ur-PK Widely spoken in care communities
Bengali bn-BD Significant diaspora population
Arabic ar-SA Growing care workforce
Chinese (Mandarin) zh-CN Service users with Mandarin preference
Cantonese yue-Hant-HK Hong Kong diaspora community
Yoruba yo-NG West African diaspora

Language is automatically selected based on service user preference and can be set per visit.

Care System Integrations

CareSignal exports structured care data for import into existing UK care management platforms:

Platform Format Use Case
Birdie JSON / CSV Direct import into Birdie care management
PASS (everyLIFE) JSON / CSV Care plan and observation sync
Generic NHS JSON / CSV Interoperable format for other providers
FHIR R4 Bundle JSON NHS Spine, GP Connect, Shared Care Records with SNOMED CT coding and UKCore profiles

The FHIR R4 export produces a valid Bundle containing Patient, Practitioner, Encounter, Observation, and ClinicalImpression resources with:

  • UKCore structure definition profiles
  • SNOMED CT concept codes for care observations
  • Confidence scores as FHIR extensions
  • CQC domain tagging per observation

The Problem

  • 595,000 home care workers in England
  • 59,000 unfilled posts (13% vacancy rate, 4x national average)
  • 30 to 48 minutes per shift lost to documentation
  • 297,500 hours of care capacity lost to paperwork every day
  • 24.7% annual staff turnover

Sources: Skills for Care 2025, HSE 2024/25

The Solution

CareSignal turns documentation from a manual task into an ambient process. Three taps, under 60 seconds: Start Visit, End Visit, Submit Note.

At scale (10% UK market penetration), CareSignal returns the equivalent of 90,000 FTE carers' time back to direct care.

Academic Foundation

Grounded in 30+ peer-reviewed papers from PubMed, medRxiv, WHO, and The Lancet across eight research pillars: ambient AI scribes, patient presence, home care admin burden, lone worker safety, NLP/LLM extraction, workforce crisis, AI ethics, and health systems frameworks.

Privacy and Security

  • Audio is ephemeral: deleted immediately after transcription
  • Transcripts are transient: used only for field extraction, then deleted
  • Only structured, carer-approved fields are persisted
  • TLS 1.3 in transit, AES-256 at rest
  • GDPR-compliant data minimisation by design
  • UK data residency (GCP europe-west2)

Team

Built at UCL Hub, London for the HSIL Harvard Health Systems Hackathon 2026.

License

MIT

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors